Project presentation

Laurine Panel, Rasmus Tøffner-Clausen, Mathias Glerup Filtenborg, Alexander Videbæk, Qiju Chen

2025-12-02

Introduction

Dataset

The dataset focuses on patients with prostate cancer : one of the most common cancer types among men
Prostate dataset from this data base : https://hbiostat.org/data/

Aim

Find out which treatment is the most effective
Determine which indicators are significant
Apply what we have learned

patno	stage	is_alive	cause_of_death	age	wt	pf	hx	sdate	rx	dtime	sbp	dbp	ekg	hg	sz	sg	ap	bm
373	3	dead	other specific non-ca	76	116	normal activity	1	1979-05-17	0.2 mg estrogen	48	13	8	heart strain	16.89844	2	9	0.3999634	0
407	4	dead	prostatic ca	78	95	normal activity	0	1978-12-06	placebo	36	15	9	heart strain	13.79883	55	10	3.1997070	0
59	3	dead	cerebrovascular	78	118	normal activity	1	1979-03-02	0.2 mg estrogen	3	16	9	heart strain	12.39844	3	9	0.3999634	0
395	4	dead	prostatic ca	81	98	normal activity	1	1978-06-25	1.0 mg estrogen	55	13	6	normal	11.19922	38	12	8.6992188	1

Materials and methods - Data preparation

Downloaded into two separate sets:

Metadata

data_dir <- "../data/"
meta_tsv_file <- "01_meta_dat_load.tsv"
if (!file.exists(str_c(data_dir, meta_tsv_file)) ) {
  prostate |>
    select(patno, stage, status, age, wt, pf, hx, sdate) |>
    write_tsv(file = str_c(data_dir, meta_tsv_file))

Treatment data

treatment_tsv_file <- "01_treatment_dat_load.tsv"
if (!file.exists(str_c(data_dir, treatment_tsv_file)) ) {
  prostate |>
  select(-stage, -status, -age, -wt, -pf, -hx, -sdate) |>
    write_tsv(file = str_c(data_dir, treatment_tsv_file))

Then converted the two raw files into a tidy dataset:
- Joined the data and meta_data
- Created numeric variables
- Renamed variables
- Removed outliers

Materials and methods

For modelling we created two augmentations.

df_long <- df_num |>
  select(c(patient_no,stage, age, weight, follow_up_months,
                        sys_blood_pressure, dia_blood_pressure,
                        serum_hemoglobin, serum_prostatic_acid_phosphatase,
                        combined_index, size_of_primary_tumor, is_alive, performance, medical_history,ekg, bone_metastases, treatment, dosage)) |> 
  pivot_longer(cols = c(stage, age, weight, follow_up_months,
                        sys_blood_pressure, dia_blood_pressure,
                        serum_hemoglobin, serum_prostatic_acid_phosphatase,
                        combined_index, is_alive, performance, medical_history,ekg, bone_metastases, treatment, dosage),
               names_to = "variable",
               values_to = "value")

augment_data <- full_df |>
  mutate(MAP = (sys_blood_pressure+ 2 *dia_blood_pressure) / 3,
         is_alive = case_when(is_alive == "alive" ~ 0,
                              is_alive == "dead" ~ 1),
         treatment_detail = case_when(
           treatment == "placebo" ~ "Placebo",
           treatment == "estrogen" & dosage == 0.2 ~ "0.2mg Estrogen",
           treatment == "estrogen" & dosage == 1.0 ~ "1.0mg Estrogen",
           treatment == "estrogen" & dosage == 5.0 ~ "5.0mg Estrogen",
           TRUE ~ NA_character_),
         treatment_detail = factor(treatment_detail, 
                                   levels = c("Placebo", 
                                              "0.2mg Estrogen", 
                                              "1.0mg Estrogen", 
                                              "5.0mg Estrogen")),)

Project presentation

Introduction

Dataset

Aim

Materials and methods - Data preparation

Metadata

Treatment data

Materials and methods

Results - Preliminary model

Results - Preliminary model

Results - Preliminary model

Results - Survival Analysis

Results - Survival Analysis

Discussion

How we can improve this project ?

Tons more data we’d like to model!

Building our own prediction model based on AP-levels.

Comparison with antigen data!